Unknown Object Detection Using a One-Class Support Vector Machine for a Cloud-Robot System

Area of Science:

• Robotics
• Computer Vision
• Artificial Intelligence

Background:

• Indoor mobile robot applications face challenges in inter-robot communication and computational power for sensor data processing.
• Existing methods struggle with degraded object recognition in complex, dynamic indoor environments, especially for multiple objects and unknown items.

Purpose of the Study:

• To present an efficient cloud-based multi-robot framework for indoor autonomous mobile robots.
• To enhance robot vision for robust object and obstacle classification using vision sensor data.
• To address limitations in recognizing unknown objects and multi-object scenes.

Main Methods:

• Developed a novel object segmentation model for separating objects in multi-object robotic views.
• Implemented a support vector data description (SVDD)-based one-class support vector machine for unknown object detection.
• Utilized a cloud-based convolutional neural network (CNN) with SoftMax for object identification and an incremental learning method for knowledge expansion.

Main Results:

• The proposed model demonstrated effective object detection and identification.
• Performance evaluation showed the model outperformed three state-of-the-art approaches.
• The system achieved enhanced usability through unknown object detection, incremental learning, and a cloud-based architecture.

Conclusions:

• The developed cloud-based framework significantly improves the capabilities of indoor mobile robots.
• The integration of advanced object recognition and unknown object detection enhances robot autonomy and applicability.
• The proposed system offers a robust solution for complex indoor robotic applications.